Cost-Aware Agents · Budget Metacognition

Frontier agents
don't know when to stop.

First surfaced in Tandemly Briefing — 2026-07-02.

Researchers at Northwestern's MLL Lab asked a deceptively direct question: can a frontier LLM agent tell when a task is going to fail before spending its full token budget? The answer is largely no. Agents waste between 28% and 64% of their allocated tokens on tasks they ultimately cannot complete, and their general task skill barely predicts this failure. BAGEN names this as a metacognitive gap that current training methods do not close.

Core finding
Budget-awareness is not a byproduct of capability. A highly capable agent is not automatically a budget-aware agent. The two skills are almost orthogonal.
scroll to explore

Agents spend first,
discover failure second.

Token budget is the one resource an agent has a finite supply of. Most agents treat it as a ceiling to hit, not a signal to read.

When an LLM agent is deployed on a real task, it is typically given a budget: a limit on how many tokens it can spend before it must stop. The budget is a practical constraint. Once spent, the task either succeeded or it did not.

The assumption baked into most agent designs is that the model will spend the budget productively. It will search, reason, call tools, and arrive somewhere useful before the counter runs out. Failure is understood as a terminal event, something discovered after the budget is exhausted.

The BAGEN paper challenges that framing. The question it asks is not "what does the agent do with its budget?" but "does the agent know whether the budget will be enough?" Those are different questions with different answers. A capable agent might be very good at the first and completely blind to the second.

The practical cost of that blindness is measurable. If an agent cannot recognize it is headed for failure, it will proceed at full pace even on tasks it cannot complete, spending the entire budget on a trajectory that was doomed from early on. That is not a small problem. It is the systematic over-allocation of compute to lost causes.

The question BAGEN asks

Can an LLM agent predict its own success probability against a given token budget, and act on that estimate before the budget runs out? Not "is the agent good at tasks?" but "does the agent know when it is going to fail this one?"

Evaluate, train, analyze:
three parts, one gap.

BAGEN is structured as an evaluate-train-analyze framework. Each component tests a different dimension of the budget-awareness problem.

The evaluation component establishes the baseline. The researchers measure wastage rate: the fraction of an agent's token budget that is consumed on tasks the agent ultimately fails. This is a simple but revealing metric. Most agent benchmarks report accuracy on tasks completed; wastage rate captures what happens on the tasks that are not.

The training component is the paper's intervention. The researchers use a combination of supervised fine-tuning and reinforcement learning to teach the agent two specific behaviors. The first is early-stop: the agent halts before exhausting the budget when its internal assessment indicates the task is unlikely to succeed. The second is alert: rather than stopping silently, the agent signals to the caller that the task may be failing and that additional budget, a different approach, or human escalation is warranted.

The analysis component investigates why capable agents are not budget-aware by default. The key instrument here is a correlation analysis between task skill (how accurately a model completes tasks when it does complete them) and budget-awareness (whether it correctly predicts which tasks will fail within a given budget). If capability predicted budget-awareness, the correlation would be high. It is not.

Standard agent design
Budget is an external ceiling. The agent runs until it succeeds or the limit is hit. No internal model of success probability. Failure is discovered at termination, not predicted before it.
BAGEN's frame
Budget is an active control signal. The agent estimates its own success probability and uses that estimate to trigger early-stop or alert before the budget runs out. Failure prediction becomes a first-class behavior.
Why this is not the same as step-level pruning

Prior cost-aware agent work, including Budget-Aware Value Trees and LaTER, focuses on detecting dead-end paths mid-execution: step-level marginal gain scores that prune low-value branches as the agent runs. BAGEN targets a different level. It asks whether the agent can estimate task-level doom before or early in execution, based on a self-assessment of success probability rather than a marginal step score. Both are useful. They address different failure windows.

28 to 64 percent,
wasted.

Three numbers define the paper's central result: wastage rate, skill-awareness correlation, and the training ceiling.

28–64%
Token wastage rate
on failing tasks
across frontier models
r = 0.35
Correlation between
task skill and
budget-awareness
47%
Interval coverage
reached via
SFT + RL training

The wastage range (28% to 64%) reflects real variation across models, but even the best performer wasted more than a quarter of its budget on tasks it could not complete. The range is not random noise. It correlates with other model properties, but the baseline problem is present across all frontier systems tested.

The skill-awareness correlation is the finding that lands hardest. r = 0.35 is a weak signal. It says that knowing how accurate a model is tells you very little about how well it predicts its own failures. A model that solves hard problems reliably does not, by that capability alone, know when this specific task is beyond its reach. The two skills are almost orthogonal.

The training result (47% interval coverage) shows the gap is closable, at least partially. Interval coverage measures how often the agent's predicted stopping point falls within an acceptable window around the true optimal stopping time. 47% is a meaningful improvement over the untrained baseline, and the authors are clear it is a floor: the training approach has headroom, and the training signal can be refined. The point is that budget-awareness is learnable, just not learned by default.

The training signal gap

Standard RLHF optimizes for task completion. The reward is positive when the agent finishes the task and negative or zero when it does not. That training signal contains no information about whether the agent should have stopped earlier. Budget prediction, the ability to estimate success probability and act on it, is a skill the optimization target does not reward. The gap between task-completion reward and budget-aware behavior is where the 28-to-64% wastage lives.

What interval coverage measures

Unlike a simple accuracy number on task completion, interval coverage captures whether the agent's predicted stopping point is close to the actual optimal stopping point. An agent that always predicts it should use 100% of its budget scores zero on interval coverage on every task it fails. An agent that learns to predict failure early and acts on that prediction scores higher. The metric penalizes both overconfidence (proceeding past the optimal stop) and underconfidence (stopping too early on tasks it would have completed).

What changes if you
take this seriously.

Budget-awareness is not a default. It is a capability to build, evaluate, and train for explicitly.

1
For agent developers: measure your wastage rate first
Log token usage separately for tasks that succeed and tasks that fail. Calculate the mean tokens spent on failing tasks as a fraction of the budget. That number is your wastage rate, and it is almost certainly higher than you assumed. The BAGEN result suggests even well-resourced frontier models run at 28% minimum. Most production agents are not frontier models with well-tuned prompts.
2
For architects: add a budget-awareness probe to your eval suite
The BAGEN findings suggest a new eval split worth running: a holdout of tasks your agent historically fails, measured by whether the agent completes the full budget or stops early. A high completion rate on a task the agent fails is direct evidence of a budget-awareness problem. This split is orthogonal to your standard accuracy eval and captures a failure mode that accuracy numbers cannot see.
3
For fine-tuners: the training signal gap is real and fixable
If you are fine-tuning agents on task-completion reward, you are not generating budget-awareness training signal. To close the gap, add examples of correct early-termination decisions to your training data: cases where the agent correctly recognized task doom and stopped or alerted rather than proceeding. Add a reward term that gives credit for efficient stopping on tasks the model is unlikely to complete. The BAGEN SFT+RL result shows this works; 47% interval coverage is not a ceiling.
4
For teams running cost-sensitive deployments: the two-mode design is a concrete pattern
Early-stop and alert are distinct behaviors with different use cases. Early-stop is appropriate when the task can be queued, retried with a different approach, or escalated without user visibility. Alert is appropriate when the user or system needs to know that the task is failing and make a routing decision. Designing these as separate behaviors in your agent loop, with separate threshold calibration, gives you finer control over cost and user experience than a single binary stop.
5
For researchers and evaluators: report wastage alongside accuracy
A benchmark result that reports accuracy on completed tasks does not tell you anything about what the agent does on the tasks it fails. Wastage rate is a second number that rounds out the picture. A system that fails fewer tasks but also wastes less budget on the ones it fails has a meaningfully different operational cost profile than one that fails the same number at full-budget spend. Both numbers belong in an evaluation table.

Where to go
from here.

Concrete next steps for building budget-awareness into an agent.

1
Instrument for wastage before anything else
Log token usage by task outcome: succeeded, failed, abandoned. Calculate the mean and distribution of tokens consumed on each outcome. Your wastage rate on failed tasks is the number to track. If you do not have this, you cannot know whether any intervention is working.
2
Add a self-assessment checkpoint to your agent loop
At defined intervals in your agent loop (after the first 20% of budget is spent, and again at 40%), prompt the agent to estimate its current success probability on the task. If the estimate falls below a calibrated threshold, trigger the early-stop or alert path. Log the checkpoint fire rate separately so you can tune the threshold against your actual wastage rate.
3
Read the paper
Lin, Wang et al. (2026). BAGEN: Are LLM Agents Budget-Aware? Northwestern MLL Lab. arXiv:2606.00198.
4
Build a failing-task holdout
Identify 20 to 30 tasks from your production distribution that your current agent reliably fails. Run your agent on them and measure: how often does it stop early versus consuming the full budget? The completion rate on failing tasks is your baseline wastage signal. Use it as a target metric for any budget-awareness intervention you add.
5
Review related work on cost-aware agent design
BAGEN sits alongside a cluster of 2025-2026 work on efficiency at different levels: LaTER (latent-space reasoning to cut token spend on individual chains), BoundaryRouter (routing queries between plain inference and full agent execution), and Budget-Aware Value Trees (step-level marginal gain pruning). Each addresses a different window of the cost problem. BAGEN addresses the task-level prediction window that the others do not cover.